Multi-Polygenic prediction of Frailty and its Trajectories highlights Chronic Pain, Rheumatoid Arthritis, and Educational Attainment pathways

Abstract

Frailty is a complex ageing-related trait with a growing evidence base for genetic influence. While a single polygenic score (PGS) for frailty has shown predictive value, few studies have examined the joint effect of multiple genetic risks. This study used a multi-polygenic score (MPS) approach to evaluate the combined and relative contributions of 26 PGSs to frailty, measured via the Frailty Index (FI), in two UK cohorts aged 65 and older: the English Longitudinal Study of Ageing (ELSA) and the Lothian Birth Cohort 1936 (LBC1936). Using elastic net regression with repeated cross-validation, we identified chronic pain and depressive symptoms PGSs as the strongest risk predictors of cross-sectional frailty status, while educational attainment, parental longevity, and rheumatoid arthritis PGSs were protective. Compared to single PGS models, MPS models provided improved prediction of frailty levels, explaining up to 4.7% of variance in frailty status - an improvement over the best single PGS (2.5%). To assess whether PGSs also predicted longitudinal frailty progression, we applied generalized additive mixed models (GAMMs) to model age-related trajectories. In ELSA, five PGSs (chronic pain, depressive symptoms, rheumatoid arthritis, educational attainment, and parental death) significantly interacted with age, influencing the rate of frailty change. In LBC1936, consistent though weaker effects were observed for chronic pain and education PGSs. These findings show that polygenic liability shapes both frailty levels and trajectories in later life. Our results support the use of multi-trait genomic models to improve risk prediction and understanding of frailty's complex aetiology.

Figure 1.

Bar plot of the significant single model PGS with the Frailty as the outcome, the y axis represents the polygenic scores and the x axis represents beta coefficients with 95% confidence intervals. The R2 of each model is displayed on the bars.

Figure 2.

Multi-polygenic prediction of the Frailty Index in ELSA (top left) and LBC1936 (top right: Wave 1; bottom left: Wave 3; bottom right: Wave 5). Bar plots show standardized beta coefficients from Elastic Net regularised regression models including selected polygenic scores (PGSs) and covariates (age, sex, and ancestry principal components). Stars indicate significance after FDR correction: p < 0.05 (), p < 0.01 (), p < 0.001 (). The R² values represent the additional variance in frailty explained by the multi-PGS predictor set beyond covariates alone. Note: Although “Age” and “Sex” are displayed on the y-axis alongside PGSs, they are covariates in the model and not polygenic scores - displayed to show relativity in effect size.

Figure 3.

Smooth trajectories from Generalised Additive Mixed Models (GAMMs) showing interaction effects between polygenic scores (PGSs) and age on frailty progression in ELSA. Models include penalised splines for age (s(age)) and interaction smooths (s(age, by = PGS)) to assess whether genetic liability modifies the rate of frailty increase with age. Plots display the estimated smooth effect of each PGS on the age–frailty slope, with 95% confidence intervals. Covariates include age, sex, and ten ancestry principal components. Only PGSs showing significant age interactions are plotted.